The best analytics and AI tools in the world can’t account for the unique foibles of human beings. Image: iStock/metamorworks After years of resisting “pretend football,” I finally joined a Continue Reading

The best analytics and AI tools in the world can’t account for the unique foibles of human beings.

Image: iStock/metamorworks

After years of resisting “pretend football,” I finally joined a neighborhood fantasy football league. I’m a very casual football fan and probably couldn’t name 10 active players without several minutes of thought, but in the interest of participating in some neighborly fun and learning a bit more about the game, I created my first team.

I frankly still don’t fully understand fantasy football scoring and all the nuances, but for the unfamiliar, you select a virtual team from a pool of available players during a draft process, and each player’s activities on the field that week contribute to your overall team score. For example, if my defense blocks a touchdown, I might get 10 points, while if a running back on my team rushes for a few yards in a different game, I get a fraction of a point. Theoretically, this creates interest in more teams by giving the fan more players to follow, but at this point, it’s mainly creating confusion as my extremely limited “football brain” attempts to follow a half dozen simultaneous games.

SEE: Digital transformation: A CXO’s guide (free PDF) (TechRepublic)

Our league uses the Yahoo! Fantasy Sports app/website, and while it has been years since I’ve used anything from Yahoo, the app and website are extremely impressive. Most notable for this rank amateur is the breadth and depth of statistics available, from those you would expect about a player’s past performance to predictions of the outcome of each matchup in our league. My first game had predicted scores for each of my players with two-decimal precision, predicted overall scores, and a victory probability, all of which were updated in real time throughout the weekend’s games.

I started the day as an underdog, but through some combination of luck and happenstance, my team has apparently won unless my kicker, who plays today, somehow scores–13 points. In studying the app at random intervals on Sunday, I couldn’t help but feel like I was looking at my stock broker’s online trading platform. Seemingly precise figures in red and green, flashing numbers, lent a veneer of digital certainty over what’s essentially a throw of some wildly complex dice.

Those messy humans

This randomness might seem highly undesirable. After all, no one wants an unpredictable outcome to major surgery, an airplane flight, expensive steak dinner, or even their neighborhood fantasy football league, where top prizes include a few cans of local beer. As tech leaders, the mitigation strategy for unpredictability is often automation, or more recently, analytics and AI.

At countless conferences and in the pages of technology books, I’ve heard of a brave new world of sorts, where machines make nearly perfect decisions and reduce or eliminate the “messiness” of humans. Of course, this is not without precedent, and machines have proved themselves capable and even superior at everything from flying fighter planes to winning complex games like Go.

However, despite real-time analytics, live data feeds, and way more processing power than my paltry novice football brain, the machines could not accurately predict the outcome of my fantasy football matchup. Not only did they miss the victory prediction, but the initial prediction had me with only a 39% chance of winning.

One might suggest that this is a wildly unfair task to expect a machine to perform correctly. After all, the outcome of any sporting event could hinge on something obvious like the weather, to something trivial like what a key athlete had for breakfast. All this is true, however, the danger for tech leaders is the implied certainty that comes from everything from visual cues like seemingly precise predictions, to the richness of data fed into a predictive model.

The analytical model that predicted my running back would score 15.89 points probably had years of high-quality data, and may have been developed by some of the best data scientists, but some combination of chance and circumstance conspired to have that player deliver a 4.90. Missing the mark by 70% is fine for fantasy football, but probably not so great for tasks from transoceanic navigation to sales forecasting.

As tech leaders, it’s our job to accurately convey what technologies like AI, analytics, and machine learning can and can’t do. These models may have uncanny and seemingly magical abilities in some areas, yet lack abilities that even a child performs with ease in others. These dichotomies become even more challenging when the majority of the users, and in the case of neural networks, even the creators or the network, can’t begin to understand how the models work. Whether you’re striving to win neighborhood bragging rights, or make a “bet the company” move into a new market, understand the tools at your disposal and the flaws and abilities of each.

Also see

ASNF